The storage cloud
for devs & agents
Bottomless object storage with zero egress fees. Feed training jobs, serve inference, and checkpoint agent state — all over a familiar S3 API.
A virtual bash environment backed by Tigris object storage.
Storage that gets out of your way
Tigris® is globally redundant storage, built for the new cloud.
Bottomless storage
No per-account limits. No per-region limits. Just scale.
Zero egress fees
Read as often as you need, from any region. No bandwidth bills.
Speaks S3 natively
Point boto3, aws-cli, rclone, PyTorch at Tigris. Swap the endpoint, keep the code.
Storage primitives for
agents and teams
The building blocks you already know from S3 — plus the ones you wish it had. Zero-copy forks, point-in-time snapshots, object events, and fine-grained IAM, built for the new cloud.
Pick single-region, multi-region, or global placement per bucket. Keep data close to users and redundant across regions without running your own replication.
Standard, infrequent, and archive tiers with automatic lifecycle rules. Same API, priced for your data's access pattern.
Fork a production bucket instantly. Test migrations, run experiments, or give each agent its own sandbox — with no storage duplication.
Whole-bucket snapshots. Roll back to any precise second to recover from a bad write, a bug, or a hallucinated agent.
Webhooks fire the moment an object is created, updated, or deleted. Wire up data pipelines without polling.
S3-compatible IAM plus per-bucket, per-prefix, and per-key policies. Scope credentials down to individual agents.
Serve buckets from your own domain with managed TLS. Keep branding consistent across public assets and CDNs.
Time-scoped, scope-limited URLs for direct upload and download. Let clients hit storage without ever touching your credentials.
Built for Agent Scale
Storage made for running millions of agents
Give each agent its own copy-on-write clone of a shared dataset. Instant at any size, zero duplication, isolated writes.
import { createForks, teardownForks } from "@tigrisdata/agent-kit";
// 'my-dataset' must have snapshots enabled
const { data: forkSet } = await createForks("my-dataset", 3, {
prefix: "experiment-run-42", // controls fork bucket names
credentials: { role: "Editor" }, // scoped keys per fork
});
// Each fork is its own bucket with isolated storage
for (const fork of forkSet.forks) {
console.log(fork.bucket);
}
// Clean up — revokes credentials, deletes all fork buckets
await teardownForks(forkSet);
Provision a dedicated bucket for a single agent, with optional TTL for auto-cleanup and scoped credentials for least-privilege access.
import { createWorkspace, teardownWorkspace } from "@tigrisdata/agent-kit";
const { data: workspace } = await createWorkspace("agent-workspace-abc", {
ttl: { days: 1 }, // auto-expire objects after 1 day
enableSnapshots: true, // allow checkpointing later
credentials: { role: "Editor" }, // scoped access key
});
// Use workspace.bucket + workspace.credentials to read/write
// with @tigrisdata/storage
// Clean up — revokes credentials, deletes bucket
await teardownWorkspace(workspace);
Capture bucket state at a point in time and restore it later. Restore creates a copy-on-write fork — the original is untouched.
import { checkpoint, restore, listCheckpoints } from "@tigrisdata/agent-kit";
// Take a checkpoint
const { data: ckpt } = await checkpoint("training-data", {
name: "epoch-50",
});
// List all checkpoints
const { data: list } = await listCheckpoints("training-data");
for (const c of list.checkpoints) {
console.log(c.snapshotId, c.name, c.createdAt);
}
// Restore into a new fork from that checkpoint
const { data: restored } = await restore("training-data", ckpt.snapshotId, {
forkName: "training-data-retry",
});
Wire up event-driven multi-agent pipelines. Bucket notifications fire webhooks on create, update, and delete — no polling.
import {
setupCoordination,
teardownCoordination,
} from "@tigrisdata/agent-kit";
// Configure webhook notifications on a bucket
await setupCoordination("pipeline-bucket", {
webhookUrl: "https://my-service.com/webhook",
filter: 'WHERE `key` REGEXP "^results/"',
auth: { token: "my-webhook-secret" },
});
// Clear notifications when the pipeline wraps up
await teardownCoordination("pipeline-bucket");
Store once, cache anywhere
Tigris Acceleration Gateway is a local caching proxy that delivers near-local NVMe throughput for AI training and inference — drop-in S3 compatible, with zero code changes. Deploy across regions and clouds and keep data next to every GPU.
Explore Acceleration GatewayThe fastest generative AI in the world — running on Tigris.
Predictable data costs for training and inference
Eliminate data transfer costs across clouds. AI teams keep GPUs fed and budgets flat — no surprise egress, no per-region math.
Migrate as you go, stay in sync
Objects are copied to Tigris on access, while all writes are mirrored back to your legacy bucket. Run both in parallel as long as you want—no deadlines, no lock-in. If anything’s off, your old bucket still has the latest data. This kind of write-through migration is unique to Tigris.
Explore Zero-Downtime MigrationStart with S3-compatible tools, or build your way
Keep everything. Swap the endpoint.
Tigris is S3-compatible. Point your existing SDK at t3.storage.dev and you're done — no new SDKs, no data migration weekend.
A modern object storage platform built for global performance, cost efficiency, and AI. Data is instantly accessible worldwide, with zero egress fees, and it plugs into the tools you already use.
When you need globally distributed storage for AI/ML workloads, want to eliminate egress fees, need S3 compatibility with existing tools, or require infinite scale without capacity planning.
Tigris is global-by-default, zero-egress, and ships purpose-built AI features like bucket forking. Unlike S3 or GCS, your data is automatically available worldwide — no cross-region copy pipelines to build.
Yes. SOC 2 Type II certified, HIPAA BAA available, with the same availability guarantees as major cloud providers. All data is encrypted at rest and in transit.
No storage limits. Tigris scales infinitely — store as much as you need without capacity planning or pre-provisioning.
